When Kaled Alhanafi’s team goes into a medical practice, it brings stopwatches. The team measures how long employees take to process a referral, reach a patient and arrange an appointment. It asks what happens when a patient doesn’t answer or misses a visit. Employees spend their days handling faxes and phone calls, with little time to examine the process. Basata, the company Kaled co-founded to automate that work with artificial intelligence, begins by measuring it.
Learning what the patient cannot see
Kaled experienced medical delays from the other end of the phone. Before Basata, his father developed a 95% carotid artery blockage and received referrals to three cardiology groups in Phoenix. According to Kaled, one took eight weeks to call, another called after his father’s surgery, and the third never responded. He spent a significant amount of time on hold, trying to establish whether a practice had actually received the referral. The family had little visibility into what stood between his father and an appointment.
Kaled had spent years at Lyft and Cruise helping bring self-driving cars onto public roads. Yet getting his father in front of a specialist had proven to be just as challenging. That frustration led him to start Basata with co-founders, Chetan Patel, who had spent 12 years at Medtronic developing pacemakers, and Vivin Paliath. They left their jobs to build the company, but Kaled recognized how much he still had to learn. He understood the delays from the patient lens, but understanding what caused them would require spending time inside the practices.
In the company’s first months, the founders went into practices and processed referrals themselves, coming home exhausted. They watched employees open faxes, review documents and manually enter information into electronic health records. Some referral packets ran to 100 pages. Once employees entered the information, someone still had to contact the patient and decide where to schedule the visit. Kaled began to see how much judgment went into what appeared, from outside the practice, to be routine administrative work.
A referral to cardiology could leave several questions unanswered. Did the patient need help with atrial fibrillation, an artery problem or a heart that was not pumping effectively? Was the referral urgent, and did it require a subspecialty? Kaled compares those ambiguities to a problem from autonomous vehicles. A traffic light says stop while a police officer signals the car forward. The system has to interpret instructions in context.
His work in autonomous vehicles also involved working with Phoenix police officers on how to stop a self-driving car. Operating on public roads required accounting for the people who would encounter the vehicle. In health care, physicians and administrators needed to trust a new system enough to use it. Kaled chose to build in Phoenix partly to stay close to them. The area’s specialty practices gave the team places to observe the work and develop the product alongside its users.
Following the referral all the way through
Basata’s first product read incoming referral documents, extracted information and entered it into the practice’s electronic health record. In one early case, Kaled says, a practice had 500 referrals in a backlog that delayed processing by two to four weeks. An AI agent could process referrals the same day. The call center could then reach patients sooner, although employees still had to make those calls. Basata added a voice agent to contact patients and schedule visits using the practice’s rules and physician preferences, then extended into appointment reminders.
That progression shaped Kaled’s resistance to building the company around a single AI capability. Reading documents and making calls had to work together, with information passing from one task to the next. He describes a scheduling agent calling with the referral already available, confirming the patient’s identity and discussing the information needed to arrange the visit. The patient would have less to repeat because the agent could draw on what the practice had already received.
Some obstacles required changes to how the practice organized its work. Basata employs people with Lean and Six Sigma backgrounds who examine processes alongside its technologists. Kaled describes showing customers the improvement they could expect from adding an AI agent, then the larger gain possible with workflow adjustments. Reducing the administrative load gives them room to consider how they would arrange the work differently.
Hiring people who have run the practice
Working through those changes required a different team from the one Kaled initially assembled. Early hiring concentrated on engineering and product, with employees focused on getting the software into the market. Implementation called for technical staff who could sit inside a clinic and communicate with its leaders and employees. Kaled describes customers with 20, 50 or 100 physicians. Engineers had to explain the system and learn from people who knew how to run a medical practice.
Basata then began hiring practice administrators. Kaled says roughly 70% of sales come through referrals and word of mouth, and some experienced operators had already helped the company reach new customers. Bringing them onto the team added people who understood the pressures customers faced. They knew what it was like to anticipate another day of phones ringing and faxes arriving. In some cases, they already knew the administrators at the practices Basata was entering.
Those relationships helped the company learn the details its software needed. Kaled describes scheduling instructions kept in binders or in employees’ heads, with different preferences for individual physicians. A new scheduler might receive a binder explaining how to book appointments for 50 doctors. Basata had to capture those rules and make them usable by its agents. How much of that work would it have to repeat at the next practice?
Making the next deployment easier
Kaled expects the rules Basata records to become useful beyond the practice where it first encounters them. Another deployment in the same specialty can draw on work the team has already completed. He compares the process to mapping roads for autonomous vehicles. The map takes effort to create, but the company does not need to rebuild it every time a car travels through the same area. Entering a new specialty still means learning another set of requirements.
He says some implementations that took months now take weeks, particularly within similar specialties. He expects the accumulated rules and relationships inside practices to make Basata harder to replicate. Better AI models can improve the software’s capabilities, while this knowledge helps determine how those capabilities work for a customer. Unfamiliar specialties still require more work on the ground.
Basata also uses its own AI agents to reduce implementation work, including reading requirements and configuring parts of the system. Kaled expects those tasks to become less laborious while customers continue to have a contact who understands health care operations. He intends to keep that support relationship as configuration gets easier. The team is trying to reduce the effort behind implementation while preserving the help customers receive.
Kaled suggests a practical test for whether a product has become useful; ask customers what would happen if you took it away. Basata encountered that test when a system upgrade briefly interrupted service and customers called asking for it to be restored immediately. His team explained that it would take a few minutes, but even that short wait prompted concern. To Kaled, those calls showed how much the practices had come to rely on Basata. An interruption meant employees faced the prospect of taking back the work they handed over.